With the rise of novel computing techniques such as big data and artificial intelligence, many scientific and industrial disciplines are faced with exponentially increasing demands for data storage and compute resources. Traditional data compression algorithms are either generically applicable but lossless, limiting performance (e.g. zip), or lossy but designed for specific applications (e.g. jpg, mp3). Machine learning can be deployed to identify the most significant features of any given dataset and favour these features in a compression algorithm. Baler is a novel framework for developing, testing and deploying autoencoder-based data compression algorithms. In this talk, we report on recent developments to the Baler framework, including the implementation of Baler on FPGAs and how Baler has been used to compress data from atomic physics (Mössbauer imaging).